Reproduce “Human2Dex: Dexterous Robot Learning from Wrist-Centric Human Demonstrations” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jiachen Zhang, Yuanyuan Li, Shaqi Luo
Human2Dex reuses bare-hand wrist-camera demonstrations to train separate policies for Linker O6 and Wuji. It aligns hand observations, retargets motion into each hand’s native commands, and represents object position relative to a grasp pocket. Across six tasks, it records 206 successes in 240 trials versus 63 for Human-Raw. Results cover tested hands and tasks, not unseen ones.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.